Reverse Hillclimbing, Genetic Algorithms and the Busy Beaver Problem
|
|
- Ethel Simmons
- 5 years ago
- Views:
Transcription
1 Reverse Hillclimbing, Genetic Algorithms and the Busy Beaver Problem Terry Jones Gregory J. E. Rawlins SFI WORKING PAPER: SFI Working Papers contain accounts of scientific work of the author(s) and do not necessarily represent the views of the Santa Fe Institute. We accept papers intended for publication in peer-reviewed journals or proceedings volumes, but not papers that have already appeared in print. Except for papers by our external faculty, papers must be based on work done at SFI, inspired by an invited visit to or collaboration at SFI, or funded by an SFI grant. NOTICE: This working paper is included by permission of the contributing author(s) as a means to ensure timely distribution of the scholarly and technical work on a non-commercial basis. Copyright and all rights therein are maintained by the author(s). It is understood that all persons copying this information will adhere to the terms and constraints invoked by each author's copyright. These works may be reposted only with the explicit permission of the copyright holder. SANTA FE INSTITUTE
2 Reverse Hillclimbing, Genetic Algorithms and the Busy Beaver Problem Terry Jones * Gregory J. E. Rawlins t April 29, 1993 Abstract This paper introduces a new analysis tool called reverse hillclimbing, and demonstrates how it can be used to evaluate the performance of a genetic algorithm. Using reverse hillclimbing, one can calculate the exact probability that hillclimbing will attain some point in a landscape. From this, the expected number of evaluations before the point is found by hillclimbing can be calculated. This figure can be compared to the average number of evaluations done by a genetic algorithm. This procedure is illustrated using the Busy Beaver problem, an interesting problem of theoretical importance in its own right. At first sight, a genetic algorithm appears to perform very well on this landscape, after examining only a vanishingly small proportion of the space. Closer examination reveals that the number of evaluations it performs to discover an optimal solution compares poorly with even the simplest form of hillclimbing. Finally, several other uses for reverse hillclimbing are discussed. 'Santa Fe Institute, 1660 Old Pecos Trail, Suite A., Santa Fe, NM 87501, USA terry@santafe.edu tdepartment of Computer Science, Indiana University, 215 Lindley Hall, Bloomington, IN 47405, USA. rawlins@cs.indiana.edu 1
3 1 Introduction 1.1 Fitness Landscapes Many problems can be viewed as a search in a space of solutions. Ifthe structure of a solution can be expressed as settings for each of n variables, one can imagine an n dimensional space with an axis for each of the variables. The neighborhood of a point P in the space is the set of all points that differ from P in exactly one variable. When each variable is binary, each point has n neighbors. These will be called P's one-mutant neighbors. If the quality of a proposed solution can be quantified, then each point in the space has associated with it a number corresponding to its worth as a solution, called its fitness. The entire space, including the fitness values, is called a fitness landscape or simply a landscape. In this paper, higher quality solutions are assumed to have numerically higher fitness values. Initially, the fitnesses of all points in the space are unknown. The task is to identify high points on the landscape, or, ideally, the highest point(s). Calculating the fitness of a point is called evaluating the point and is done by an evaluation function. A search algorithm is judged by the number of points it evaluates as well as the quality of the solutions it finds. Exhaustive search will always find the optimal solution, but the space is usually too large to make this an acceptable search method. At the other end of the spectrum, randomly choosing a single point requires just one evaluation, but quality is unlikely to be high. Between these extremes are many alternatives, including random search, genetic algorithms and the many forms of hillclimbing (such as gradient descent and simulated annealing). Different problems have different landscapes, so an algorithm may be good on one type of landscape but not on another. Further, an algorithm may perform well in one region of a space and not in another. In addition, the probabilistic nature of most of these algorithms, makes it hard to predict what they will do from run to run. All of these issues make it hard to compare the performance of different algorithms. It is worthwhile to study smaller, better understood spaces and the performance of algorithms on them. There are many questions that we can answer, to some extent, on these landscapes: questions about specific algorithms, comparisons between algorithms, and questions about the nature of the landscapes themselves. 1.2 Genetic Algorithms Genetic Algorithms, (GAs) were first proposed by Holland [7]. A GA maintains a population of points on the landscape and uses various methods modeled on genetics to evolve new, and hopefully fitter, populations. The ideas that are most commonly borrowed from genetics are mutation and recombination. Each solution is viewed as the genome of an individual in the population (see [6] for further references). 2
4 1.3 Hillclimbing and Reverse Hillclimbing Hillclimbing typically chooses a random starting point in the landscape then attempts to move uphill to better one-mutant neighbors, until it can no longer do so. In Steepest Ascent Hillclimbing all the one-mutant neighbors of the current point are examined. If there are fitter neighbors, move to the fittest. If there are several equally fit best neighbors, choose one at random. To properly evaluate a hillclimber both in relation to other hillclimbers and in,elation to GAs it is important to answer the question: What is the probability that a hillclimb, started from a random point in the space, will discover a given point P? This question was formerly an extremely difficult one to answer on a rugged landscape. Reverse hillclimbing answers it very quickly and exactly. In fact, the more rugged and difficult to search a landscape is, the better is the performance of reverse hillclimbing. Formerly, it was at worst computationally infeasible and at best slow and inaccurate to even approximate this probability via repeated random hillclimbs. The question can be answered by locating all the points in P's basin of attraction and the probability that each will hillclimb to P. These probabilities can then be summed and the total divided by the size of the space. Reverse hillclmibing does this by descending from P in all possible directions, but only to points that could possibly re-ascend to P. The algorithm has at heart a recursive function reverse...hillclimbo that takes as arguments a point P and a probability p that the point hillclimbs to P. Initially, the function is called with some peak P and p = 1. A global hash table keeps track of all the points that have been encountered, their probabilities of climbing to P, and a pointer to a list of their interesting descendants (see the second and third items below). The function does the following: 1. If P is not in the hash table, add it and its probability p, otherwise, increment the existing probability by p. 2. Make a list L of all P's one-mutant neighbors that are less fit than P. This can be obtained from the hash table if P has been seen before. 3. For each element Q of L, check to see if Q hillclimbs back to P, and if so, determine the probability, q that it does so. Recurse, calling reverse...hillclimb(q, pq). When the recursion is finished, it is simple to calculate the probability that a single steepest ascent hillclimb from a randomly chosen point in the space will reach P. The following two questions can now be answered: Given that a GA has located a certain peak, how long, on average, would it take a hillclimbing algorithm to find the same peak, and how does this compare with the time the GA takes? The reciprocal of the probability that reverse hillclimbing produces is the expected number of hillclimbs it will take until P is encountered. Using this number and an estimate, obtained through sampling, of the average uphill walk length, one obtains the expected number of evaluations before P is found. This number can be compared directly with the 3
5 k ~(k) 8(k) ': 4,098 2': 47,176,870 6 > 95,524,079 > 8,690,333,381,690,952 Table 1: The State of The Busy Beaver World observed performance of the GA, thus giving information that can be used to better tune the GA or to determine whether the GA is doing useful work. 1.4 The Busy Beaver Problem In the Busy Beaver problem [13] deterministic Turing Machines (TMs) with alphabet {O, I} and a fixed number of states k (plus a halt state) are sought that write as many 1 symbols as possible on an initially zero-filled two-way infinite tape before halting. The number of 1 symbols that remain on the tape is the machine's score. The problem is to find, for a given k, the machine with the highest score, a number denoted by ~(k). The number of steps the machine makes before halting is called the shift number and is denoted 8(k). More than one machine may generate ~(k) ones and these machines do not necessarily have the highest shift numbers amongst all k-state machines. The problem is intractable. The function defined by ~(k) is not computable, it grows faster than any computable function. Even k = 2 is difficult to do by hand. Thirty years ago Rado [12] wrote: "Even though skilled mathematicians and experienced programmers attempted to evaluate ~(3) and 8(3), there is no evidence that any known approach will yield the answer, even if we avail ourselves of high-speed computers and elaborate programs. As regards ~(4), 8(4) the situation seems to be entirely hopeless at present." Today we know that his pessimism was justified; exact answers are known only for k ::::: 4. These proofs have been developed as much through concerted human effort as through bruteforce computer enumeration [2, 3]. See [3] and [9] for details of k = 4 and 5 respectively. The figures for k = 6 in Table 1 have not yet been published or independently verified [10J. As further illustration of the growth of ~(k), Schult (see [5]) found a 12-state machine that generates 6 * ones. The ellipsis in the exponent stack represents 163 missing exponents, each of 4,096. The number of possible machines on k states is, in the simplest enumeration, (4 +4k )2k. In the range of interest, 3 ::::: k ::::: 9, this number increases by three orders of magnitude for each increment of k. Further, the difficulty of the problem is greatly compounded by 4
6 the extreme growth in the shift number. From Table 1 it can be seen that the shift number increases by a factor of over 400,000 as k increases from 4 to 5, and by over 180,000,000 when k moves from 5 to 6. This is important since testing machines may require simulating them for this many more steps. Without further theoretical insight, to discover a new 5-state champion, requires simulating machines for at least fifty million steps. The connection with the halting problem should be apparent. Search for Busy Beavers in the past has been in two directions: finding faster ways to simulate TMs, and finding smarter ways to prove that machines do not halt. The first has seen elaborate software contrivances and even the construction of special hardware for running TMs. The second approach involves either static analysis of the TM's transition rules or dynamic analysis of the TM's behavior. So far exhaustive search combined with fast simulation and good inferencing has been the source of all the best champions found. 2 Searching For Busy Beavers With A GA 2.1 Coding and Parameters We used a character-based encoding. A TM with k states was represented by a character string of 6k bytes. When the machine is in a certain state, looking at a certain symbol, it needs to know three things: the next state, the symbol to write on the tape and the direction in which to move. Since the machine could be scanning either a 0 or a 1, six bytes are required. By restricting crossover to byte boundaries, only legal TM's are generated. For the results reported below, the following GA parameters were used: tournament selection (of size 2) with the higher fitness individual selected with probability 0.75, mutation at a rate of 0.01 (a mutation always causes an allele change), two point crossover with probability 0.75, 50 generations, and a' generation gap of 0.9 (the surviving 10% were the best from the generation). The population size varied with k. We used populations as large as 2,000 for k = 4. Approximately twenty different parameter settings were tried and the above settings did as well as any. 2.2 Results In all the cases where the optimal score is known (k ::; 4), the GA found a machine that achieves it. We have not attempted runs for k = 5 due to the large number of simulation steps it is apparently necessary to run machines for in order to find a new champion. The results are summarized in Table 4. For k = 4 we were encouraged to find that the GA discovered an optimal solution after examining, on average, only 0.7% of the entire space, and this with a naive TM simulator. But how did the GA find the optimal machine? Was it luck? Was this performance good? Disabling crossover produced a marked decrease in performance. Was crossover finding and using good schema? Attempting to answer these questions led to reverse hillclimbing. 5
7 k Size of Space E(k) Number of Peaks , ,777, ,600,000, Table 2: The Number of Peaks of Optimal Fitness for k ::; 4 k P(Locate) Climbs Walk Length Neighbors Evals (0) 1.00 (0.71) (25) 1.33 (0.85) (220) 1.61 (1.02) 30 10, ,129 (497,129) 1.87 (1.15) 48 44,741,100 Table 3: Expected Evaluations to Find Optimal TMs Using Steepest Ascent 3 What Reverse Hillclimbing Can Tell Us With no more than the machines found by the GA in hand, we could have made progress with reverse hillclimbing. But, happily, the problem has a thirty year history and we were able to draw on the work of others who had found the peaks years earlier. It is important to note, however, that reverse hillclimbing does not need to first know the highest peaks, it can be used with any set of points and any space to derive information both about the space and about the probable performance of hillclimbing algorithms on that space. Thus, the following comparisons would be valid even if we only had information discovered through using only a GA. The landscapes for k ::; 4 were examined in some detail. Once again, k = 5 is not considered here as nothing, apart from lower bounds, can be said about it with certainty. Table 2 shows the number of optimal peaks in these four landscapes. Table 3 shows the results of reverse hillclimbing from these peaks. The second column shows the probability that a single hillclimb, started from a randomly chosen point in the space, will reach an optimal peak. This figure is calculated via reverse hillclimbing, and is exact (in the decimal places shown). The third column shows the expected number of hillclimbs until a peak is found, with standard deviations in parentheses. This is just the reciprocal of the probability of a successful ascent, this being the expected value of a geometric random variable. The fourth column gives the average length of a steepest ascent climb on the landscape (found via sampling). Again, standard deviations are given in parentheses. The fifth column shows the number of neighbors that must be examined in each step of a steepest ascent. This is given by 2k(k + 2). The final column gives the 6
8 k GA Evals Hillclimb Evals 1 8 (9.27) (4.245) 2 8,117 (8,663) 542 (508) 3 123,404 (128,022) 10,606 (10,492) 4 186,666,666 (89,449,934) 42,372,351 (19,123,501) Table 4: The GA Versus Steepest Ascent Hillclimbing expected number of evaluations until a peak is found - this is the product of columns three and four and five. For the k = 4 case, reverse hillclimbing found the probability after evaluating only 161,748 points out of 25,600,000,000 (0.0006% of the space). This figure would be very hard to obtain with any accuracy via sampling with iterated hillclimbing, which would only arrive at an optimal peak approximately once in a million evaluations, and of course for the Busy Beaver problem, 'evaluation' is often not cheap. We can now compare these expectations against the performance of the GA. 4 Comparing The G A With A Hillclimber Table 4 shows the average number of evaluations taken by the GA and by a steepest ascent hillclimber. Standard deviations follow the means. Note the close correspondence between the hillclimbing results and the predicted results in Table 3. The hillclimber is finding peaks for k = 4 about four and a half times as fast as the GA. This does not necessarily imply that every GA would be worse than a hillclimber on this problem. The main point here is that we can use reverse hillclimbing to quickly and accurately assess how easily a hillclimber would find a result that a GA found. This is easily compared to the GA's performance and from this comparison one may be content with the GA, modify GA parameters, or even change search algorithms. 5 Discussion Reverse hillclimbing has several other potential applications. Firstly, some observations: Any hillclimber can be reversed if it satisfies two conditions. The first is that it must be possible to calculate the algorithm's next possible moves and their probabilities. Secondly, the algorithm cannot be one which moves to a point whose fitness is less than or equal to the current point. Hillclimbers which violate the second condition can potentially loop indefinitely, and must be analyzed mathematically. These conditions allow us to perform reverse hillclimbing on all of Steepest Ascent, Next Ascent, Random Ascent, First Ascent, Random Bit Climbing ([4]), and various forms 7
9 of mutational hillclimbing, but not on Stochastic Hillclimbing ([1]), Bit Setting ([14]), Random Mutation Hillclimbing ([11]) or any of the forms of simulated annealing. One can do a roughly equivalent thing in the uphill direction, a probabilistic ascent, and find all the peaks that could be ascended to from P, each with a probability. There is a trade-off between how useful reverse hillclimbing will be and how rugged the landscape is. In the cases where the landscape is not rugged, hillclimbing can answer some questions quickly that reverse hillclimbing cannot, and vice-versa. If a peak has a small basin of attraction, reverse hillclimbing can be very useful. Hillclimbing algorithms vary in terms of the amount of work done before a move is made and also in how deterministic they are. Less determinism generally means that a point can climb to more peaks, but each with smaller probability. Greater determinism means that peaks have smaller basins of attraction, but once a climb is started within one, the probability of reaching the peak is higher. These tradeoffs make it difficult to compare two hillclimbing algorithms both in general and for a specific landscape. However with reverse hillclimbing, one can make direct comparisons between hillclimbing algorithms on a landscape. By sampling points, or using peaks found by hillclimbing or some other search, and reverse hillclimbing from them with different hillclimbers, one can tell which hillclimber will on average do better. Using reverse hillclimbing and the probabilistic ascent mentioned above, one can obtain a measure of how a system in equilibrium, a spin glass for example, that is then disturbed is likely to settle, and how much energy is needed to bump the system into a new configuration. Reverse hillclimbing establishes the basin of attraction and probabilistic ascent then identifies the peaks that could be returned to and their probabilities. If a landscape has peaks that are close together, as has been observed in N f{ landscapes [8], one could use reverse hillclimbing with probabilistic ascent to find the peaks that are close to a given peak. Reverse hillclimbing can establish the entire basin of attraction, and probabilistic ascent can then identify all peaks that can be ascended to from all these points. Any peak found using this method is 'close' to the original in the sense that there is at least one saddle point between them. Iterating this procedure could quickly identify small mountainous regions of the space. This is essentially a new search method that will work on spaces that have this characteristic. The two methods can be used to arrive at statistical properties of a fitness landscape. For instance, the average number of points "below" and "above" a given point, or some measure relating the number of uphill directions to the number of peaks that can be reached. Reverse hillclimbing could also be used in immune system modeling to determine the basin of attraction for a given antigen. One could then determine the set of existing antibodies that could climb to the antigen and their probabilities. Reverse hillclimbing could identify antigens that would not elicit an immune response, so-called "holes in the immune system" - this would happen if the model of the body had no antibody in the antigen's basin of attraction. 8
10 6 Conclusion This paper introduced a new tool for landscape analysis, reverse hillclimbing and showed one way it can be used to examine the performance of a GA. The performance of a GA on the Busy Beaver problem was examined and further suggestions were made for the use of reverse hillclimbing and probabilistic ascent. 7 Acknowledgements Many thanks to Allen Brady and Heiner Marxen who were very helpful and willing to share their knowledge and latest results on the Busy Beaver problem, despite the deluge of questions. Thanks too to Stuart Kauffman for his thoughts on the wider applications of reverse hillclimbing. References [1] Ackley, David H. [1987] A Connectionist Machine For Genetic Hillclimbing, Kluwer Academic Publishers. [2] Brady, A. H. [1966], "The Conjectured Highest Scoring Machines for Rado's 2;(k) for the Value k = 4" IEEE Trans. on Electronic Computers vol. EC-15, pp [3] Brady, A. H. [1983J, "The Determination of the Value of Rado's Noncomputable Function 2;(k) for Four-State Turing Machines" Mathematics of Computation, vol. 40, no April 1983 pp [4] Davis, Lawrence [1989] "Bit-Climbing, Representational Bias, and Test Suite Design" In Richard K. Belew and Lashon B. Booker (Eds.) Proceedings of the Fourth International Conference on Genetic Algorithms, Morgan Kaufmann, San Mateo, CA, pp [5] Dewdney, A. K. [1989] "Noncomputable Functions - The Busy Beaver Problem" The Turing Omnibus, Computer Science Press, Ch. 36, pp [6] Goldberg, David E. [1989], Genetic Algorithms in Search, Optimization and Machine Learning, Addison-Wesley, [7] Holland, John H. [1975J, Adaptation in Natural and Artificial Systems, University of Michigan Press, [8] Kauffman, Stuart A. [1989], "Adaptation on Rugged Fitness Landscapes" Lectures in the Sciences ofcomplexity, Ed. D. Stein, Addison-Wesley Longman, vol. 1, pp [9J Marxen, H. and Buntrock, J. [1990], "Attacking The Busy Beaver 5" Bulletin of The European Association for Theoretical Computer Science, vol. 40, pp
11 [10] Marxen, H. [1992]. Personal communication. [11] Palmer, Richard. [1992]. Personal communication. - [12] Rado, T. [1963], "On a Simple Source for Non-computable Functions." Proceedings of the Symposium on Mathematical Theory ofautomata, Polytechnic Institute of Brooklyn, April 1963, pp [13] Rado, T. [1962], "On Non-computable Functions" Bell System Technical Journal, vol. 41, pp [14] Wilson, Stewart W. [1991] "GA-Easy Does Not Imply Steepest-Ascent Optimizable" In Richard K. Belew and Lashon B. Booker (Eds.) Proceedings ofthe Fourth International Conference on Genetic Algorithms, Morgan Kaufmann, San Mateo, CA, pp
Busy Beaver The Influence of Representation
Busy Beaver The Influence of Representation Penousal Machado *, Francisco B. Pereira *, Amílcar Cardoso **, Ernesto Costa ** Centro de Informática e Sistemas da Universidade de Coimbra {machado, xico,
More information(Anti-)Stable Points and the Dynamics of Extended Systems
(Anti-)Stable Points and the Dynamics of Extended Systems P.-M. Binder SFI WORKING PAPER: 1994-02-009 SFI Working Papers contain accounts of scientific work of the author(s) and do not necessarily represent
More informationEvolutionary Computation
Evolutionary Computation - Computational procedures patterned after biological evolution. - Search procedure that probabilistically applies search operators to set of points in the search space. - Lamarck
More informationRepresentation and Hidden Bias II: Eliminating Defining Length Bias in Genetic Search via Shuffle Crossover
Representation and Hidden Bias II: Eliminating Defining Length Bias in Genetic Search via Shuffle Crossover Abstract The traditional crossover operator used in genetic search exhibits a position-dependent
More informationLocal Search & Optimization
Local Search & Optimization CE417: Introduction to Artificial Intelligence Sharif University of Technology Spring 2017 Soleymani Artificial Intelligence: A Modern Approach, 3 rd Edition, Chapter 4 Outline
More informationTuring s thesis: (1930) Any computation carried out by mechanical means can be performed by a Turing Machine
Turing s thesis: (1930) Any computation carried out by mechanical means can be performed by a Turing Machine There is no known model of computation more powerful than Turing Machines Definition of Algorithm:
More informationLocal Search & Optimization
Local Search & Optimization CE417: Introduction to Artificial Intelligence Sharif University of Technology Spring 2018 Soleymani Artificial Intelligence: A Modern Approach, 3 rd Edition, Chapter 4 Some
More informationLOCAL SEARCH. Today. Reading AIMA Chapter , Goals Local search algorithms. Introduce adversarial search 1/31/14
LOCAL SEARCH Today Reading AIMA Chapter 4.1-4.2, 5.1-5.2 Goals Local search algorithms n hill-climbing search n simulated annealing n local beam search n genetic algorithms n gradient descent and Newton-Rhapson
More informationImplicit Formae in Genetic Algorithms
Implicit Formae in Genetic Algorithms Márk Jelasity ½ and József Dombi ¾ ¾ ½ Student of József Attila University, Szeged, Hungary jelasity@inf.u-szeged.hu Department of Applied Informatics, József Attila
More informationComputational statistics
Computational statistics Combinatorial optimization Thierry Denœux February 2017 Thierry Denœux Computational statistics February 2017 1 / 37 Combinatorial optimization Assume we seek the maximum of f
More informationOn the Speed of Quantum Computers with Finite Size Clocks
On the Speed of Quantum Computers with Finite Size Clocks Tino Gramss SFI WORKING PAPER: 1995-1-8 SFI Working Papers contain accounts of scientific work of the author(s) and do not necessarily represent
More informationA Statistical Genetic Algorithm
A Statistical Genetic Algorithm Angel Kuri M. akm@pollux.cic.ipn.mx Centro de Investigación en Computación Instituto Politécnico Nacional Zacatenco México 07738, D.F. Abstract A Genetic Algorithm which
More informationGenetically Generated Neural Networks II: Searching for an Optimal Representation
Boston University OpenBU Cognitive & Neural Systems http://open.bu.edu CAS/CNS Technical Reports 1992-02 Genetically Generated Neural Networks II: Searching for an Optimal Representation Marti, Leonardo
More informationSearch. Search is a key component of intelligent problem solving. Get closer to the goal if time is not enough
Search Search is a key component of intelligent problem solving Search can be used to Find a desired goal if time allows Get closer to the goal if time is not enough section 11 page 1 The size of the search
More informationRenormalization Group Analysis of the Small-World Network Model
Renormalization Group Analysis of the Small-World Network Model M. E. J. Newman D. J. Watts SFI WORKING PAPER: 1999-04-029 SFI Working Papers contain accounts of scientific work of the author(s) and do
More informationMost General computer?
Turing Machines Most General computer? DFAs are simple model of computation. Accept only the regular languages. Is there a kind of computer that can accept any language, or compute any function? Recall
More informationA.I.: Beyond Classical Search
A.I.: Beyond Classical Search Random Sampling Trivial Algorithms Generate a state randomly Random Walk Randomly pick a neighbor of the current state Both algorithms asymptotically complete. Overview Previously
More informationTuring Machines Part III
Turing Machines Part III Announcements Problem Set 6 due now. Problem Set 7 out, due Monday, March 4. Play around with Turing machines, their powers, and their limits. Some problems require Wednesday's
More informationGenetic Algorithms and Genetic Programming Lecture 17
Genetic Algorithms and Genetic Programming Lecture 17 Gillian Hayes 28th November 2006 Selection Revisited 1 Selection and Selection Pressure The Killer Instinct Memetic Algorithms Selection and Schemas
More informationThe Role of Crossover in Genetic Algorithms to Solve Optimization of a Function Problem Falih Hassan
The Role of Crossover in Genetic Algorithms to Solve Optimization of a Function Problem Falih Hassan ABSTRACT The genetic algorithm is an adaptive search method that has the ability for a smart search
More informationMore Turing Machines. CS154 Chris Pollett Mar 15, 2006.
More Turing Machines CS154 Chris Pollett Mar 15, 2006. Outline Multitape Turing Machines Nondeterministic Turing Machines Enumerators Introduction There have been many different proposals for what it means
More informationLocal Search (Greedy Descent): Maintain an assignment of a value to each variable. Repeat:
Local Search Local Search (Greedy Descent): Maintain an assignment of a value to each variable. Repeat: I I Select a variable to change Select a new value for that variable Until a satisfying assignment
More information22c:145 Artificial Intelligence
22c:145 Artificial Intelligence Fall 2005 Informed Search and Exploration III Cesare Tinelli The University of Iowa Copyright 2001-05 Cesare Tinelli and Hantao Zhang. a a These notes are copyrighted material
More informationScaling Up. So far, we have considered methods that systematically explore the full search space, possibly using principled pruning (A* etc.).
Local Search Scaling Up So far, we have considered methods that systematically explore the full search space, possibly using principled pruning (A* etc.). The current best such algorithms (RBFS / SMA*)
More informationLocal Search and Optimization
Local Search and Optimization Outline Local search techniques and optimization Hill-climbing Gradient methods Simulated annealing Genetic algorithms Issues with local search Local search and optimization
More informationAttacking the Busy Beaver 5
Attacking the Busy Beaver 5 Heiner Marxen, Specs GmbH, Berlin Ju.. rgen Buntrock, Technische Universitȧ. t Berlin ABSTRACT Since T. Rado in 1962 defined the busy beaver game several approaches have used
More informationLocal Beam Search. CS 331: Artificial Intelligence Local Search II. Local Beam Search Example. Local Beam Search Example. Local Beam Search Example
1 S 331: rtificial Intelligence Local Search II 1 Local eam Search Travelling Salesman Problem 2 Keeps track of k states rather than just 1. k=2 in this example. Start with k randomly generated states.
More informationComputational Models Lecture 8 1
Computational Models Lecture 8 1 Handout Mode Ronitt Rubinfeld and Iftach Haitner. Tel Aviv University. April 18/ May 2, 2016 1 Based on frames by Benny Chor, Tel Aviv University, modifying frames by Maurice
More informationWhat Makes a Problem Hard for a Genetic Algorithm? Some Anomalous Results and Their Explanation
What Makes a Problem Hard for a Genetic Algorithm? Some Anomalous Results and Their Explanation Stephanie Forrest Dept. of Computer Science University of New Mexico Albuquerque, N.M. 87131-1386 Email:
More informationTuring machines Finite automaton no storage Pushdown automaton storage is a stack What if we give the automaton a more flexible storage?
Turing machines Finite automaton no storage Pushdown automaton storage is a stack What if we give the automaton a more flexible storage? What is the most powerful of automata? In this lecture we will introduce
More informationCausal Effects for Prediction and Deliberative Decision Making of Embodied Systems
Causal Effects for Prediction and Deliberative Decision Making of Embodied ystems Nihat y Keyan Zahedi FI ORKING PPER: 2011-11-055 FI orking Papers contain accounts of scientific work of the author(s)
More informationPart I: Definitions and Properties
Turing Machines Part I: Definitions and Properties Finite State Automata Deterministic Automata (DFSA) M = {Q, Σ, δ, q 0, F} -- Σ = Symbols -- Q = States -- q 0 = Initial State -- F = Accepting States
More informationQ = Set of states, IE661: Scheduling Theory (Fall 2003) Primer to Complexity Theory Satyaki Ghosh Dastidar
IE661: Scheduling Theory (Fall 2003) Primer to Complexity Theory Satyaki Ghosh Dastidar Turing Machine A Turing machine is an abstract representation of a computing device. It consists of a read/write
More informationHaploid-Diploid Algorithms
Haploid-Diploid Algorithms Larry Bull Department of Computer Science & Creative Technologies University of the West of England Bristol BS16 1QY, U.K. +44 (0)117 3283161 Larry.Bull@uwe.ac.uk LETTER Abstract
More informationCS 331: Artificial Intelligence Local Search 1. Tough real-world problems
S 331: rtificial Intelligence Local Search 1 1 Tough real-world problems Suppose you had to solve VLSI layout problems (minimize distance between components, unused space, etc.) Or schedule airlines Or
More informationComplex Systems Theory and Evolution
Complex Systems Theory and Evolution Melanie Mitchell and Mark Newman Santa Fe Institute, 1399 Hyde Park Road, Santa Fe, NM 87501 In Encyclopedia of Evolution (M. Pagel, editor), New York: Oxford University
More informationGenetic Algorithm. Outline
Genetic Algorithm 056: 166 Production Systems Shital Shah SPRING 2004 Outline Genetic Algorithm (GA) Applications Search space Step-by-step GA Mechanism Examples GA performance Other GA examples 1 Genetic
More informationCrossover Gene Selection by Spatial Location
Crossover Gene Selection by Spatial Location ABSTRACT Dr. David M. Cherba Computer Science Department Michigan State University 3105 Engineering Building East Lansing, MI 48823 USA cherbada@cse.msu.edu
More information1 Showing Recognizability
CSCC63 Worksheet Recognizability and Decidability 1 1 Showing Recognizability 1.1 An Example - take 1 Let Σ be an alphabet. L = { M M is a T M and L(M) }, i.e., that M accepts some string from Σ. Prove
More informationINVARIANT SUBSETS OF THE SEARCH SPACE AND THE UNIVERSALITY OF A GENERALIZED GENETIC ALGORITHM
INVARIANT SUBSETS OF THE SEARCH SPACE AND THE UNIVERSALITY OF A GENERALIZED GENETIC ALGORITHM BORIS MITAVSKIY Abstract In this paper we shall give a mathematical description of a general evolutionary heuristic
More informationCS 21 Decidability and Tractability Winter Solution Set 3
CS 21 Decidability and Tractability Winter 2018 Posted: January 31 Solution Set 3 If you have not yet turned in the Problem Set, you should not consult these solutions. 1. (a) A 2-NPDA is a 7-tuple (Q,,
More informationEVOLUTIONARY OPERATORS FOR CONTINUOUS CONVEX PARAMETER SPACES. Zbigniew Michalewicz. Department of Computer Science, University of North Carolina
EVOLUTIONARY OPERATORS FOR CONTINUOUS CONVEX PARAMETER SPACES Zbigniew Michalewicz Department of Computer Science, University of North Carolina Charlotte, NC 28223, USA and Thomas D. Logan IBM, Charlotte,
More informationData Warehousing & Data Mining
13. Meta-Algorithms for Classification Data Warehousing & Data Mining Wolf-Tilo Balke Silviu Homoceanu Institut für Informationssysteme Technische Universität Braunschweig http://www.ifis.cs.tu-bs.de 13.
More informationCPSC 467b: Cryptography and Computer Security
CPSC 467b: Cryptography and Computer Security Michael J. Fischer Lecture 10 February 19, 2013 CPSC 467b, Lecture 10 1/45 Primality Tests Strong primality tests Weak tests of compositeness Reformulation
More informationThe Bootstrap is Inconsistent with Probability Theory
The Bootstrap is Inconsistent with Probability Theory David H. Wolpert SFI WORKING PAPER: 1995-10-091 SFI Working Papers contain accounts of scientific work of the author(s) and do not necessarily represent
More informationThe concept of the adaptive landscape
1 The concept of the adaptive landscape The idea of a fitness landscape was introduced by Sewall Wright (1932) and it has become a standard imagination prosthesis for evolutionary theorists. It has proven
More informationThe Evolution of Sex Chromosomes through the. Baldwin Effect
The Evolution of Sex Chromosomes through the Baldwin Effect Larry Bull Computer Science Research Centre Department of Computer Science & Creative Technologies University of the West of England, Bristol
More information(Preliminary Version) Evolutionary Turing Machines The Quest for Busy. Beavers
(Preliminary Version) Evolutionary Turing Machines The Quest for Busy Beavers Penousal Machado, Francisco B. Pereira Instituto Superior de Engenharia de Coimbra (ISEC), Quinta da Nora, 3030 Coimbra, Portugal.
More informationComputational Models Lecture 8 1
Computational Models Lecture 8 1 Handout Mode Ronitt Rubinfeld and Iftach Haitner. Tel Aviv University. May 11/13, 2015 1 Based on frames by Benny Chor, Tel Aviv University, modifying frames by Maurice
More informationDevelopment. biologically-inspired computing. lecture 16. Informatics luis rocha x x x. Syntactic Operations. biologically Inspired computing
lecture 16 -inspired S S2 n p!!! 1 S Syntactic Operations al Code:N Development x x x 1 2 n p S Sections I485/H400 course outlook Assignments: 35% Students will complete 4/5 assignments based on algorithms
More informationSample Project: Simulation of Turing Machines by Machines with only Two Tape Symbols
Sample Project: Simulation of Turing Machines by Machines with only Two Tape Symbols The purpose of this document is to illustrate what a completed project should look like. I have chosen a problem that
More informationEvolutionary Design I
Evolutionary Design I Jason Noble jasonn@comp.leeds.ac.uk Biosystems group, School of Computing Evolutionary Design I p.1/29 This lecture Harnessing evolution in a computer program How to construct a genetic
More informationSmall Turing machines and generalized busy beaver competition
Small Turing machines and generalized busy beaver competition Pascal MICHEL Équipe de logique de l université Paris 7 and IUFM de l académie de Versailles michel@logique.jussieu.fr January, 2004 Abstract
More informationAn Analysis of Diploidy and Dominance in Genetic Algorithms
An Analysis of Diploidy and Dominance in Genetic Algorithms Dan Simon Cleveland State University Department of Electrical and Computer Engineering Cleveland, Ohio d.j.simon@csuohio.edu Abstract The use
More informationIntroduction to Algorithms
Lecture 1 Introduction to Algorithms 1.1 Overview The purpose of this lecture is to give a brief overview of the topic of Algorithms and the kind of thinking it involves: why we focus on the subjects that
More informationA Genetic Algorithm for Optimization Design of Thermoacoustic Refrigerators
Proceedings of the 7th WSEAS International Conference on Simulation, Modelling and Optimization, Beijing, China, September 15-17, 2007 207 A Genetic Algorithm for Optimization Design of Thermoacoustic
More informationFinding an Impulse Response Function Using Genetic Programming
Finding an Impulse Response Function Using Genetic Programming Martin A. Keane John R. Koza James P. Rice Third Millennium Venture Capital Limited 5733 Grover Chicago, Illinois 663 32-777524 Computer Science
More informationDetermination of Optimal Tightened Normal Tightened Plan Using a Genetic Algorithm
Journal of Modern Applied Statistical Methods Volume 15 Issue 1 Article 47 5-1-2016 Determination of Optimal Tightened Normal Tightened Plan Using a Genetic Algorithm Sampath Sundaram University of Madras,
More informationThe Story So Far... The central problem of this course: Smartness( X ) arg max X. Possibly with some constraints on X.
Heuristic Search The Story So Far... The central problem of this course: arg max X Smartness( X ) Possibly with some constraints on X. (Alternatively: arg min Stupidness(X ) ) X Properties of Smartness(X)
More informationGENETIC ALGORITHM FOR CELL DESIGN UNDER SINGLE AND MULTIPLE PERIODS
GENETIC ALGORITHM FOR CELL DESIGN UNDER SINGLE AND MULTIPLE PERIODS A genetic algorithm is a random search technique for global optimisation in a complex search space. It was originally inspired by an
More informationLocal and Online search algorithms
Local and Online search algorithms Chapter 4 Chapter 4 1 Outline Local search algorithms Hill-climbing Simulated annealing Genetic algorithms Searching with non-deterministic actions Searching with partially/no
More informationComputational Models Lecture 8 1
Computational Models Lecture 8 1 Handout Mode Nachum Dershowitz & Yishay Mansour. Tel Aviv University. May 17 22, 2017 1 Based on frames by Benny Chor, Tel Aviv University, modifying frames by Maurice
More informationA Tractable Walsh Analysis of SAT and its Implications for Genetic Algorithms
From: AAAI-98 Proceedings. Copyright 998, AAAI (www.aaai.org). All rights reserved. A Tractable Walsh Analysis of SAT and its Implications for Genetic Algorithms Soraya Rana Robert B. Heckendorn Darrell
More informationNEUROEVOLUTION. Contents. Evolutionary Computation. Neuroevolution. Types of neuro-evolution algorithms
Contents Evolutionary Computation overview NEUROEVOLUTION Presenter: Vlad Chiriacescu Neuroevolution overview Issues in standard Evolutionary Computation NEAT method Complexification in competitive coevolution
More informationTheory of Computation
Theory of Computation Unit 4-6: Turing Machines and Computability Decidability and Encoding Turing Machines Complexity and NP Completeness Syedur Rahman syedurrahman@gmail.com Turing Machines Q The set
More informationAutomata Theory (2A) Young Won Lim 5/31/18
Automata Theory (2A) Copyright (c) 2018 Young W. Lim. Permission is granted to copy, distribute and/or modify this document under the terms of the GNU Free Documentation License, Version 1.2 or any later
More informationCS 361 Meeting 26 11/10/17
CS 361 Meeting 26 11/10/17 1. Homework 8 due Announcements A Recognizable, but Undecidable Language 1. Last class, I presented a brief, somewhat inscrutable proof that the language A BT M = { M w M is
More information7.1 Basis for Boltzmann machine. 7. Boltzmann machines
7. Boltzmann machines this section we will become acquainted with classical Boltzmann machines which can be seen obsolete being rarely applied in neurocomputing. It is interesting, after all, because is
More informationAppearances Can Be Deceiving!
Appearances Can Be Deceiving! Overview: Students explore the relationship between angular width, actual size, and distance by using their finger, thumb and fist as a unit of angular measurement in this
More informationBusch Complexity Lectures: Turing s Thesis. Costas Busch - LSU 1
Busch Complexity Lectures: Turing s Thesis Costas Busch - LSU 1 Turing s thesis (1930): Any computation carried out by mechanical means can be performed by a Turing Machine Costas Busch - LSU 2 Algorithm:
More informationTheory of Computation Lecture Notes. Problems and Algorithms. Class Information
Theory of Computation Lecture Notes Prof. Yuh-Dauh Lyuu Dept. Computer Science & Information Engineering and Department of Finance National Taiwan University Problems and Algorithms c 2004 Prof. Yuh-Dauh
More informationEvolutionary Computation. DEIS-Cesena Alma Mater Studiorum Università di Bologna Cesena (Italia)
Evolutionary Computation DEIS-Cesena Alma Mater Studiorum Università di Bologna Cesena (Italia) andrea.roli@unibo.it Evolutionary Computation Inspiring principle: theory of natural selection Species face
More informationEvolutionary computation
Evolutionary computation Andrea Roli andrea.roli@unibo.it DEIS Alma Mater Studiorum Università di Bologna Evolutionary computation p. 1 Evolutionary Computation Evolutionary computation p. 2 Evolutionary
More informationPROBLEM SOLVING AND SEARCH IN ARTIFICIAL INTELLIGENCE
Artificial Intelligence, Computational Logic PROBLEM SOLVING AND SEARCH IN ARTIFICIAL INTELLIGENCE Lecture 4 Metaheuristic Algorithms Sarah Gaggl Dresden, 5th May 2017 Agenda 1 Introduction 2 Constraint
More informationLecture 14: Recursive Languages
Lecture 14: Recursive Languages Instructor: Ketan Mulmuley Scriber: Yuan Li February 24, 2015 1 Recursive Languages Definition 1.1. A language L Σ is called recursively enumerable (r. e.) or computably
More informationThe Evolution of Gene Dominance through the. Baldwin Effect
The Evolution of Gene Dominance through the Baldwin Effect Larry Bull Computer Science Research Centre Department of Computer Science & Creative Technologies University of the West of England, Bristol
More informationLecture 15: Genetic Algorithms
Lecture 15: Genetic Algorithms Dr Roman V Belavkin BIS3226 Contents 1 Combinatorial Problems 1 2 Natural Selection 2 3 Genetic Algorithms 3 31 Individuals and Population 3 32 Fitness Functions 3 33 Encoding
More informationGradient Descent. Sargur Srihari
Gradient Descent Sargur srihari@cedar.buffalo.edu 1 Topics Simple Gradient Descent/Ascent Difficulties with Simple Gradient Descent Line Search Brent s Method Conjugate Gradient Descent Weight vectors
More informationECS 120 Lesson 18 Decidable Problems, the Halting Problem
ECS 120 Lesson 18 Decidable Problems, the Halting Problem Oliver Kreylos Friday, May 11th, 2001 In the last lecture, we had a look at a problem that we claimed was not solvable by an algorithm the problem
More informationIntroduction to Turing Machines. Reading: Chapters 8 & 9
Introduction to Turing Machines Reading: Chapters 8 & 9 1 Turing Machines (TM) Generalize the class of CFLs: Recursively Enumerable Languages Recursive Languages Context-Free Languages Regular Languages
More informationAbout the relationship between formal logic and complexity classes
About the relationship between formal logic and complexity classes Working paper Comments welcome; my email: armandobcm@yahoo.com Armando B. Matos October 20, 2013 1 Introduction We analyze a particular
More informationCS4026 Formal Models of Computation
CS4026 Formal Models of Computation Turing Machines Turing Machines Abstract but accurate model of computers Proposed by Alan Turing in 1936 There weren t computers back then! Turing s motivation: find
More informationOptimizing Stochastic and Multiple Fitness Functions
Optimizing Stochastic and Multiple Fitness Functions Joseph L. Breeden SFI WORKING PAPER: 1995-02-027 SFI Working Papers contain accounts of scientific work of the author(s) and do not necessarily represent
More information(a) Definition of TMs. First Problem of URMs
Sec. 4: Turing Machines First Problem of URMs (a) Definition of the Turing Machine. (b) URM computable functions are Turing computable. (c) Undecidability of the Turing Halting Problem That incrementing
More informationMetaheuristics and Local Search
Metaheuristics and Local Search 8000 Discrete optimization problems Variables x 1,..., x n. Variable domains D 1,..., D n, with D j Z. Constraints C 1,..., C m, with C i D 1 D n. Objective function f :
More informationChapter 8: Introduction to Evolutionary Computation
Computational Intelligence: Second Edition Contents Some Theories about Evolution Evolution is an optimization process: the aim is to improve the ability of an organism to survive in dynamically changing
More informationTuring Machines. Lecture 8
Turing Machines Lecture 8 1 Course Trajectory We will see algorithms, what can be done. But what cannot be done? 2 Computation Problem: To compute a function F that maps each input (a string) to an output
More informationTuring Machines (TM) Deterministic Turing Machine (DTM) Nondeterministic Turing Machine (NDTM)
Turing Machines (TM) Deterministic Turing Machine (DTM) Nondeterministic Turing Machine (NDTM) 1 Deterministic Turing Machine (DTM).. B B 0 1 1 0 0 B B.. Finite Control Two-way, infinite tape, broken into
More informationOptimization Methods via Simulation
Optimization Methods via Simulation Optimization problems are very important in science, engineering, industry,. Examples: Traveling salesman problem Circuit-board design Car-Parrinello ab initio MD Protein
More informationFundamentals of Semiconductor Devices Prof. Digbijoy N. Nath Centre for Nano Science and Engineering Indian Institute of Science, Bangalore
Fundamentals of Semiconductor Devices Prof. Digbijoy N. Nath Centre for Nano Science and Engineering Indian Institute of Science, Bangalore Lecture - 05 Density of states Welcome back. So, today is the
More informationA Note on Turing Machine Design
CS103 Handout 17 Fall 2013 November 11, 2013 Problem Set 7 This problem explores Turing machines, nondeterministic computation, properties of the RE and R languages, and the limits of RE and R languages.
More informationExploration of population fixed-points versus mutation rates for functions of unitation
Exploration of population fixed-points versus mutation rates for functions of unitation J Neal Richter 1, Alden Wright 2, John Paxton 1 1 Computer Science Department, Montana State University, 357 EPS,
More informationMetaheuristics and Local Search. Discrete optimization problems. Solution approaches
Discrete Mathematics for Bioinformatics WS 07/08, G. W. Klau, 31. Januar 2008, 11:55 1 Metaheuristics and Local Search Discrete optimization problems Variables x 1,...,x n. Variable domains D 1,...,D n,
More information1 Computational problems
80240233: Computational Complexity Lecture 1 ITCS, Tsinghua Univesity, Fall 2007 9 October 2007 Instructor: Andrej Bogdanov Notes by: Andrej Bogdanov The aim of computational complexity theory is to study
More informationFundamentals of Metaheuristics
Fundamentals of Metaheuristics Part I - Basic concepts and Single-State Methods A seminar for Neural Networks Simone Scardapane Academic year 2012-2013 ABOUT THIS SEMINAR The seminar is divided in three
More informationWe prove that the creator is infinite Turing machine or infinite Cellular-automaton.
Do people leave in Matrix? Information, entropy, time and cellular-automata The paper proves that we leave in Matrix. We show that Matrix was built by the creator. By this we solve the question how everything
More informationAutomata Theory CS S-12 Turing Machine Modifications
Automata Theory CS411-2015S-12 Turing Machine Modifications David Galles Department of Computer Science University of San Francisco 12-0: Extending Turing Machines When we added a stack to NFA to get a
More informationIntroduction to Evolutionary Computation
Introduction to Evolutionary Computation 1 Evolutionary Computation Different lines of investigation in simulated evolution GA (Genetic Algorithms) ES (Evolution Strategies) EP (Evolutionary Programming)
More informationFundamentals of Genetic Algorithms
Fundamentals of Genetic Algorithms : AI Course Lecture 39 40, notes, slides www.myreaders.info/, RC Chakraborty, e-mail rcchak@gmail.com, June 01, 2010 www.myreaders.info/html/artificial_intelligence.html
More informationCOMPARATIVE ANALYSIS ON TURING MACHINE AND QUANTUM TURING MACHINE
Volume 3, No. 5, May 2012 Journal of Global Research in Computer Science REVIEW ARTICLE Available Online at www.jgrcs.info COMPARATIVE ANALYSIS ON TURING MACHINE AND QUANTUM TURING MACHINE Tirtharaj Dash
More information